Blind Separation Using Second Order Statistics for Non-stationary Signals1

In the signal processing area, blind source separation (BSS) is a method aiming to recover independent sources from their linear instantaneous mixtures without resorting to any prior knowledge, such as mixing matrices and sources. There have been increased attentions given to blind source separation in many areas, including wireless communication, biomedical imaging processing, multi-microphone array processing, and so on in recent years. In this paper, we propose a new simple BSS technique that exploits second order statistics for nonstationary sources. Our technique utilizes the algebraic structure of the signal model and the subspace structures so as to efficiently recover sources with interference of noise. Computer simulations have demonstrated that, in comparison with other existent methods, our method has better performance in the regimes of low and medium SNRs. For high SNRs, our method is not as promising methods such as the method called AC (“alternating columns”)-DC (“diagonal centers") algorithm, but it gives reasonable performance.